Scholar: Mitra Nasimi
When: Monday, December 1st, 3:00pm-5:00pm
Where: KH A549 (Lincoln)
Zoom: https://unl.zoom.us/j/98556596833
Title: Automated Tornado-Induced Treefall and Root Ball Detection Through Deep Learning and Analysis
Abstract: Tornadoes are among the most destructive weather phenomena, frequently occurring in the United States and leaving distinct treefall patterns that reflect the near-surface wind field in forested regions. Treefall methods use these patterns to estimate tornado intensity and characteristics. However, current treefall-based approaches are mainly labor-intensive, time-consuming, and subjective, making them difficult to apply consistently at a large scale. This research integrates deep learning, image processing, geometric, and geospatial analysis to automate the detection and geometric feature extraction for both treefalls and root balls.
A key outcome of this work is TreEAID (Treefall Evaluation, Analysis, and Identification through Deep Learning), an open-source software platform. It is the first tool designed to efficiently and automatically analyze tornado effects on forested areas using aerial imagery. The platform estimates orientation vectors and geometric features, which are used to generate wind-direction and damage maps for evaluating tornado intensity using treefall-based methods.